SIAM: A parameter-free, Spatial Intersection Attention Module

Gaoge Han, Shaoli Huang, Fang Zhao, Jinglei Tang

Abstract: Highlights•The paper introduces a novel, plug-and-play 3-D attention module without requiring parameters.•A parameter-free method simultaneously associates both spatial and channel information.•Combination of horizontal and vertical queries to form the attention map, which is simple and easy to implement.•The straightforward use of matrix products and average pooling operations makes inference speeds faster than prior methods.•Excellent performance improvements over other attention methods while obtaining higher FPS.